Technically, I should have done this as my first course and I also keep recommending others to do this course, so why not do it myself.
I will probably not write much in the notebook below as it's supposed to be more hands-on course and the result will show on my resume instead.

• Resume is your marketing tool.
• First step is automated review like Applicant Tracking System(ATS).
• Main challenge is to create a resume both for ATS and human
• This course is NOT ABOUT making AI write your resume. It's helping you.
• Five guiding principles:
  1. Privacy
  2. Prototype
  3. Structured process
  4. Focused Context
  5. Review
• ATS searches for keywords so it's important to change wordings to match it.

• LM Studio, GPT4ALL and Ollama are some ways to run local models.
• For 8GB RAM, look for a model that is 4GB or less.
• 4B models takes usually about 2-3GB.
• For a macbook with M1/M2 a 4B parameters model run fairly well.

• Reasoning Model	Non-Reasoning Model
  Outputs "thinking" text before actual output(Might not have been made visible to users)	Directly outputs its completion.
  Better results on more difficult tasks.
  	Faster Speed.

• Context Window - Maximum amount of text the model can consider. Some LLMs summarize older chat and still move forward, some forget older chats.
  Limited by RAM.
• Increasing the context window too much can cause the computer to freeze or force a restart.
• There is an effective context window which is less than the "physical context window"
• Stick with default 4096 tokens unless you have more than 8GB memory.
• Avoid sending sensitive private information for public chatbots.
• If you have to use a public chatbot, use a private LLM to remove sensitive information and then share it to public.
  Note: Always cross-check
• Hybrid mode of using both local and remote LLMs are pretty good way to balance this.
• 1. Decompose a hard task that local LLM can't do well.
  2. Public chatbot performs sub-task that doesn't require private information.
  3. Intermediate output that helps the local LLM perform it's task.
  + (1→ 3)
  4. Local LLM combines output from public chatbot and private content to perform a sub-task that is easier than the original task.
• Use your judgement to decide what you want to share or not.
• There might be sensitive information in source documents.
• One by one processing: Eating elephant one bite at a time is better.